Arao AI
Back to search results

gemma-4-E2B-it · Q3_K_M

Start a discussion
Apple M1 Prollama.cpp v0.44L2 Cross-framework verified

This page aggregates 1 real-world runs of gemma-4-E2B-it (Q3_K_M) on Apple M1 Pro with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

141.09 tok/s

Decode

Decode speed

627.28 tok/s

Prefill

Prefill speed

2.36 s

TTFT

Time to first token

4.96 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

2 days ago

Last verified

Performance

  1. llama.cpp · Q3_K_M (current)Decode 141.09 · Prefill 627.28 ·
  2. llama.cpp · Q6_K Decode 93.78 · Prefill 304.74 ·
  3. Ollama · Q3_K_L Decode 126.39 · Prefill 536.59 ·
  4. Ollama · Q5_K_M Decode 87.78 · Prefill 394.72 ·

Core figures

Decode (avg)
141.09 tok/s
Prefill (avg)
627.28 tok/s
TTFT (avg)
2.36 s
VRAM (avg)
4.96 GB
MTP acceptance rate
—
TTFB
1.13 GB
Power draw
65.6 W

Configuration

Member-level fields are taken from the most recent run

Model
gemma-4-E2B-it
Quantization
Q3_K_M
Framework
llama.cpp
Version
v0.44
Context length
4096 tokens
Batch size
512
GPU layers
33
Flash Attention
On

Hardware

Nominal and measured figures are shown side by side; whether it runs is the reader's call

GPU
Apple M1 Pro
Measured VRAM (avg)
4.96 GB
OS
macOS 15.0
Driver
Apple M1 Pro
CUDA
N/A
Power draw
65.6 W

Sources & evidence

1 measured records in total, each traceable to its original source

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-16

    v0.44 · macOS 15.0 · CUDA N/A · 4096 ctx

    141.09 tok/s

    Decode

    627.28 tok/s

    Prefill

    2.36 s

    TTFT

    4.96 GB

    VRAM

    —

    MTP

    65.6 W

    Power draw

    L2 batch seed 2026-10-16: llama.cpp on hw_id=100 running model_id=66 at Q3_K_M